Infer image shape from model for onnx export

[simonschoelly]

What has changed and why?

This PR infers the height and width from the model for the ONNX export if they re not specified.

I also infer the number of channels now from the model. I don't think it makes sense at all that one can specify the number of channels for the ONNX export - so I decided to remove that option at all.

I don't think we even have the ability to train with something else than RGB channels - so I did not write a test for grayscale images for example - let me know if that is actually possible.

How has it been tested?

(Delete this: Please describe the tests that you ran to verify your changes. Provide instructions so we can reproduce. Please also list any relevant details for your test configuration.)

Did you update CHANGELOG.md?

• Yes
• Not needed (internal change)

Probably not necessary

Did you update the documentation?

• Yes
• Not needed (internal change without effects for user)

[Copilot]

Pull Request Overview

This PR modifies the ONNX export functionality to infer image dimensions from the trained model instead of requiring explicit specification. The key changes remove the hardcoded channel parameter and make height/width optional by inferring them from the model's configuration.

• Removes the num_channels parameter and infers channel count from the model's normalization configuration
• Makes height and width parameters optional, defaulting to the model's trained image size
• Updates test cases to cover scenarios with None values for height/width dimensions

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

File	Description
src/lightly_train/_commands/export_task.py	Modified export functions to infer image dimensions and channels from model, updated parameter types and documentation
tests/_commands/test_export_task.py	Added test cases with None values for height/width and logic to use model's image_size when dimensions are None

[yutong-xiang-97]

Approved.